La búsqueda desesperada por potencia informática: por qué van a seguir subiendo los precios de los móviles por la IA

The relentless pursuit of AI supremacy has sparked a fierce competition for computing power, driving up the prices of smartphones. The unexpected success of Anthropic's Claude Code in early 2024 revealed the demand for computational capacity, leading to service outages. To address this, Anthropic's co-founder Tom Brown visited Elon Musk's offices in California in late March and secured a 45 billion dollar agreement for over 300 megavolts of computational power over the next 45 years.

This was not the first time Anthropic had sought such partnerships; they had previously spoken to Meta's AI director Alexandr Wang about securing more computational capacity, but Meta declined the offer. Anthropic's revenue from Claude Code alone jumped to $6 billion annually in February, and they would have added more capacity if it were available.

As demand for computational power grows, it is becoming increasingly scarce, pushing companies to form alliances and agreements with rivals to secure a slice of the limited market. The four major US hyperscalers (Amazon, Google, Meta, and Microsoft) invested $217 trillion in AI infrastructure in 2024 and $380 trillion in 2025, with predictions reaching $725 trillion by 2026.

This surge in spending is due to the lack of supply unable to meet the demand for computational power. GPU prices continue to rise, with a 60% increase in a year for a two-year Nvidia H100 GPU rental contract. Cloud providers have shifted from monthly rentals to multi-year contracts with significant upfront payments and credit checks.

Ashish Nadkarni, IDC's global enterprise infrastructure research director, explains that this situation presents a "Fear of Missing Out" (FOMO) among the major cloud providers, feeling that failure to build this capacity could render them irrelevant in the long run. However, supply chain limitations are further exacerbated by inflated demand.

The scarcity of computing power stems from several interlinked problems, with silicon shortage being the primary focus, directly affecting consumers' wallets. TSMC, a Taiwanese company producing over 80% of advanced AI chips, is facing challenges in silicon production, particularly for N3 logic wafers. Since this year, all major AI accelerators have transitioned to N3, and AI will consume most of the manufacturing capacity in 2026 and 86% in 2027.

Building capacity requires constructing new data centers, which will take two years, leaving a supply-demand gap. This tension forces TSMC to allocate production and prioritize AI chips over smartphones and PCs. Another contributing factor is the memory HBM, stacked alongside the chip and required for it to function. SK Hynix, Samsung, and Micron manufacture this memory, consuming three to four times more wafer material per bit than conventional memory.

Each gigabyte of HBM reduces capacity for mobile and computer memory. The Center for Strategic and International Studies calculates that AI centers will consume about 70% of all global memory production, but they will only satisfy around 60% of global demand. Micron has committed over 75% of its 2027 production capacity and expects tight supplies until 2030.

Samsung's production increased by 125% in the third quarter of 2026, while Micron's surged by 379%, while SK Hynix saw a 259% increase in its second quarter. The Bernstein research firm reports a 20% increase in memory chip contracts' prices in Q3 2026 compared to the previous quarter, and PC and smartphone sales have dropped to record lows.

Amidst this complex situation, a new player, the neoclouds, has emerged, purchasing GPUs and other chips and renting them out to those unable to meet their computational needs.